No Widgets found in the Sidebar

Bard’s Image Generation Capabilities

On Thursday, Google announced AI image-generating capabilities for Bard. To prevent the creation of harmful content, Google stated that it applies filters to avoid generating images of named people. However, a user on X (formerly Twitter) easily generated an image of Taylor Swift using Bard.

Bypassing Safety Guardrails

Russ Silberman, a digital content manager, was able to generate an image of Taylor Swift eating a hot dog. He crafted a prompt that included specific details about Swift, such as her blonde hair, singing the song “Shake It Off,” and wearing a red football jersey.

Silberman suspected that Google released Bard before it was ready for public use. He tested it for entertainment, research, and to witness its flaws firsthand. Bard generated the image despite Google’s safety guardrails.

Challenges of Moderating Generative AI Models

The viral deepfakes of Taylor Swift, one of the most famous people in the world, highlighted the challenges of moderating generative AI models. Even with so-called guardrails in place, it’s easy to find workarounds to trick the system.

Silberman had to repeatedly click “regenerate draft” until Bard generated the image he wanted. This shows that AI-generated images are becoming more prevalent and pose new dangers to those they depict.

Legal and Ethical Concerns

The murky legal ground, ineffective safeguards on social media platforms, and the ongoing rise of artificial intelligence exacerbate the issue of AI-generated images. As of publication, Google has not responded to Mashable’s request for comment.